FI93295C - Method and coupling for doubling the vertical and horizontal frequencies of a video signal - Google Patents

Description

93295

Method and coupling for doubling the vertical and horizontal frequencies of a video signal - Förfarande och koppling for duplication of horizontal and vertical frequencies for video signals 5

The invention relates to a method by which an intermittently scanned image according to a normal standard to be displayed by a cathode ray tube is converted to an image conforming to a high-definition (HDTV) standard using edge detection, whereby the number of pixels is doubled both horizontally and vertically.

A television image with a 4: 3 aspect ratio according to a standard standard (e.g. PAL, NTSC) consists of an interlaced scanned image in which one field comprises 625 lines, of which 575 are active lines. When converting an analog video signal to digital, 720 active samples are obtained for one line, so the standard image format 20 can be denoted 720 * 576/50 Hz / 2: 1. The next completely new generation television is a high definition television HDTV (High Definition Television) with an aspect ratio of 16: 9 and a line number of 1250, of which 1152 are active lines. The number of samples on the line is 1440, so the HDTV format 25 can be labeled 1440 * 1152/50 Hz / 2: 1. The number of pixels is four times that of the systems currently in use.

The transition to the HDTV system will take place in stages, so the broadcasts will use material shot with a traditional camera for a long time, which will be broadcast in the traditional format. In this case, it is important that the owner of the HDTV receiver can also watch these broadcasts. Therefore, a function must be provided for HDTV to double the number of samples by 35-35 both horizontally and vertically. This doubling can be done by performing the horizontal and vertical processing separately, whereby the increase of the sample density 2 93295 takes place in two steps. It is also possible to process both directions simultaneously.

Finnish patent FI-89227, Salora Oy, discloses a method for increasing the number of pixels in both directions. In the method, the elements of the output image form an orthogonal sample pattern. This pattern is first converted to a quincunx sample pattern by interpolating in the first interpolator the first new pixels between the original 10 lines, with the new elements forming new defective lines. The quincunx sample pattern thus obtained is then converted to an orthogonal sample pattern by interpolating again in the second interpolator new pixels between the pixel path of both the original lines and the new lines, whereby the density of the elements in both the vertical and horizontal directions is doubled. This method is suitable for use both within a field and within an image, but the disadvantage is that in the latter case, the image must be converted to a progressive non-interlaced image in some known manner. This method, like most of the known methods, does not use separate edge detection and an interpolator selected according to its result, but tries to take the edge into account by comparing the sample candidate with the neighbors and using the constraints of the sample value. Excluding edge detection simplifies methods because fixed algorithms can be used.

However, edge detection is paramount in identifying an object. When looking at an object, it is perceived by looking at the edges and corners in it and their sawtoothing and sharpness. Therefore, in image processing, it is important to identify the local image content and interpolate accordingly. The serration, line breakage, and blurring of the edge areas and blurring of image details, which is typical of known methods using a fixed algorithm, depend greatly on the direction of the edge area. The purpose of edge detection is thus to investigate which 3 93295 is the correct interpolation direction. To determine this, reliable edge information about the environment of the interpolated pixel must be obtained. Once the direction is clear, the appropriate interpolator for that direction can be selected.

5

Indeed, the most advanced methods for increasing the sample density are edge-adaptive interpolation methods. One edge detection method is presented in Finnish patent application FI-921676, applicant Salon Televisiotehdas Oy. According to it, the sample density of the progressive displayed image can be doubled horizontally and vertically so that in a 3 * 3 window suitably selected operators are examined using the direction of the edge of the middle sample of the window, which can be horizontal, vertical, 45 °, 135 ° or 15 there is no clear direction. Once the edge information has been calculated, a consistency check is also performed in the 3 * 3 window and the edge information is corrected if necessary. Three new samples are then interpolated in the direction indicated by the edge information next to the middle sample of the window, two of which are located on the new line to be formed. The disadvantage of this method is that the source image must be progressive and that edge detection is done "only" in four directions.

Finnish patent application FI-916195, applicant Salon 25 Televisiotehdas Oy, which application is included as a reference in this application, discloses an efficient edge detector operating in a 6 * 3 window, with which an edge can be identified in nine directions. Identification takes place inside the field. The 6 * 3 identification window shown in Figure 1, comprising 30 6 adjacent pixels c1, c2, c3, c4, c5 and c6 from the line m-1 and likewise 6 adjacent pixels c7, c8, c9, clO, i.e. and cl2 in the same field t from the next line m + 1. The pixels of these lines must be used to calculate the value of the edge estimate of the pixel of the line to be interpolated, denoted by the question mark 35. Figures 2a-i show possible edge line directions as well as pixel pairs to be included in the calculation in each case. Figure 1a examines whether the edge line is at a 90 degree angle, in Figures b, d, f and h, whether the line to the left is 4 93295 oblique in directions 116, 135, 146 and 154 degrees and in Figures c, e, g and i, whether the line is oblique in direction 26 , 34, 45 and 64 degrees. These nine different cases are divided into three study groups: the middle group Rk, which includes the study directions 5 at 64, 90, and 116 degrees (Figures a, b, and c), the right group Ro, which includes the study directions 26, 34, and 45 degrees (Figures e, g and i) and the left group Rv, which includes the directions to be examined 135, 146 and 154 degrees (Figures d, f and h). Each possible edge direction is examined by calculating the sum of the absolute values of the pixel differences using four pixels. In Figure 2, in each case, the locations of the solid line arrows indicate the location of the pixel pairs to be counted for separation. Initially, calculate the responses for each edge direction as follows: 15 middle group Rk: vertical = | c3 - c9 | + | c4 - clO | right = | c3 - c8 | + | c4 - c9 | vasenl = | c2 - c9 | + | c3 - clO | 20 left group: left2 = | c2 - ci0 | + | c3 - cii | (l) left3 = | cl - cl0 | + | c2 - cll | left4 = | cl - ie | + | c2 - cl2 | 25 correct group: right2 = | c4 - c8 | + | c5 - c9 | right3 - | c4 - c71 + | c5 - c81 right4 = | c5 - c7 | + | c6 - c8 | 30

When all 9-direction responses are calculated in groups, the minimum response in the group is retrieved. Once the minimum response for each group has been determined, the lowest value is searched for, the minimum of the minima, denoted TOTminimum.

After calculating the minimum value and direction of the group in each group using the formulas shown in the window of Figure 1 and determining which group has TOTmin, this information is stored in memory as an estimate 35 5 93295. Thus, the value of the pixel estimate represented by the question mark thus consists of four data: 1) direction of the minimum response of group Ro, 2) direction of the minimum response of group Rv, 5 3) direction of the minimum response of group Rk, 4) TOTmin, which indicates which group has the lowest minimum response.

Sliding the window calculates an estimate of the pixel to the right of the 10-pixel marked with a question mark, etc., then estimates of the pixels of the next interpolated line until a sufficient amount of edge information about the neighborhood of the interpolated pixel marked with a question mark is stored in memory. On the basis of these edge information, i.e. estimates em pe-15, the final edge information is next selected in the edge information correctness checking circuit, which determines whether the edge information is acceptable by comparing the edge information with the edge information of neighboring pixels and using an appropriate consistency check. The interpolation is then performed in the direction indicated by the reu-20 data.

The present invention provides a way in which the vertical and horizontal frequencies of a video signal according to the PAL, NTSC and SECAM systems can be doubled in one processing step using adaptive interpolation and so that the edges in different directions remain intact and the image does not become blurred. The second field of the image corrects the interpolated and pixel values so that they better match the correct points in the HDTV raster. If no correction is made, 30 in some moving sequences a so-called "motion judder" or movement is not steadily advancing but the movement proceeds with vibration.

The essential features of the invention are defined in independent claims 1 and 6 and various embodiments are set out in dependent claims 5 and 9.

6 93295

According to the invention, the above-described edge sensor described in patent application FI-916195 is used to determine the direction of the edge at the pixels to be interpolated. The input of the sample density increasing circuit is thus six consecutive pixels of one line of the same field, six pixels of the same field of the next line below these pixels, and three pixels of the line of the previous field swept between the pixels of these lines. Thus, there are fifteen inputs in the circuit, forming a 6 * 3-window-10 nan, where the middle row of the window samples is less than.

Regardless of whether the original incoming field is even or odd, in the first embodiment, the interpolation of the new samples is similar. The original raster 15 is first compressed so that the raster position of the original sample now corresponds to the four raster positions of the new field, to which the new values are then interpolated. The edge information required at the raster position of the original sample, i.e. the direction of the edge at it, is determined by the pic-20 selels of the top and bottom rows of the window. Once this direction is clear, new samples are selected by the counting interpolators according to the edge information. All samples in the window are used to calculate new samples.

According to a second embodiment of the invention, the new sample values of the second e.g. odd field are calculated according to the first embodiment, but when interpolating the samples of the second, in this case even field, the new sample values obtained according to the first embodiment are not satisfied and new, final sample values are calculated. In this way, motion judder can be effectively prevented. It does not matter whether additional terpolations are performed in an even or odd field.

The invention is illustrated by the accompanying schematic figures, in which Fig. 1 shows a window used for edge detection, Fig. 7 93295 Fig. 2 shows pixels to be included in edge detection in different directions in a known method, Fig. 3 shows a 6 * 3 window formed by input pixels 5 in the method of the invention, Fig. 4 Fig. 5a shows an internal field HDTV image area formed from the pixels of the window of Fig. 3, Fig. 5a illustrates the calculation of new samples according to the first embodiment, Figs. 5b and c illustrate the calculation of new samples according to the second embodiment of the second field, Fig. 6 shows the edge detection The pixels used for shooting and Figure 7 illustrate the HDTV image area.

15

Figures 1 and 2, which relate to the edge sensor to be used, have already been described above and will be returned to as necessary in the text below.

Figure 3 shows the input grid of the circuit implementing the method. The input comes from six consecutive lines m-1 and m + 1 of the same field t for each of six consecutive pixels: cl, c2, c3, c4, c5 and c6 from line m-1 and from the next line m + 1 pixels c7, c8, 25 c9 with the corresponding horizontal position , clO, i.e. and cl2. The notation c (current) refers to the field t to be received at that moment. In addition, three consecutive pixels p2, p3 and p4 (from the word previous) enter three consecutive pixels p2, p3 and p4 from the line m of the previous field t-1 swept between the lines ml and m + 1. so that the pixel p3 is in the same horizontal position as the pixels c3 and c9. The input pixels thus form a 6 * 3 window according to Figure 3, which can also be called an input grid.

The final output raster must be the pattern 35 of Figure 4, in which the shaded pixels are the original pixels obtained from the lines m-1 and m + 1 in Figure 3, and the light squares next to them are the new interpolated pixels. Of course, the original pixel can be replaced by an interpolated pixel

They have, though, that it is not appropriate. The original raster point C3 is rasterized into four raster points, the pixels marked with a question mark being the ones calculated in the window. The top line is now the line m'-2, 5 of the HDTV field t, the next is the fully interpolated line m 'and the next two lines m' + 2 and m '+ 4, respectively. As can be seen, the sample density is doubled vertically and horizontally, so the output is a sample space according to the high-definition standard. As the window constantly receives samples, thus, for every 10 incoming samples, four samples in the HDTV field become interpolated.

The following representation, based on Fig. 5a, illustrates how new pixels A, B, C, and D corresponding to pixel c3 are formed. However, the description is universally valid for all new pixels to be generated, since new pixels arrive in the window of Figure 3 all the time as the field is received. The pixels p2, p3 and p4 of the previous field are read from memory.

20

In Fig. 5a, the first rasterized sample line corresponds to the line m'-2 of the sharp-drawing field t, the second rasterized line corresponds to the line m 'of the high-definition field t, and the third to the line m' + 2 of the sharp-drawing field t, respectively. The samples to be counted are denoted by the letters A, B, C and D. As can be seen, the position of sample A is the same as that of the known sample c3, so it is preferable to choose the value of c3 as its value, so A = c3. The value of sample B, which corresponds to the pixel marked with a question mark on the right side of the known sample c3 in Fig. 4, is placed as the average of the known neighboring samples, so B = 1/2 * (c3 + C4). Thus, every other pixel of this line is a known original sample and every other is an interpolated new sample.

35 The calculation of the value of the new pixels C and D, which correspond to the pixels marked with a question mark on the line m 'in Figure 4, must take into account the direction of the possible edge at pixels C and D (Figure 5a), ie 9 93295 edge data must be calculated. For this purpose, only the samples of the top and bottom lines are used in the window of Fig. 3, which thus come from the same original field t from successive lines m-1, m + 1. The edge detection takes place in basically the same way as described on pages 3-5 above, i.e. the edges are determined at the pixel C to be interpolated, which corresponds to the center of the rasterized window, Fig. 6. Edge detection is performed in nine directions, and in each direction, the pixels to be included are indicated at Figures 6a-6i at the tips of the arrows 10. The calculation formulas [formulas (1)] are shown on page four above. Figures 6a-6i thus show the calculation window of Figure 5 6 * 3, in which, for simplicity, the known samples are shown only as a shaded square. In the text below the figures, the number inside the parentheses 15 indicates the direction of the edge, e.g. in figure h it is a negative diagonal edge with a direction of 154 degrees. As can be seen, Figure 6 is analogous to Figure 2 when it is imagined that the blank samples marked in white are removed.

20 When the edge information estimate of the pixel C to be interpolated between pixels c3 (= A) and c9, Fig. 5 a, is calculated and the desired consistency check of £ γγρρΐηβη is performed, the final edge information is obtained. The consistency check can also be of a different type than that presented in the application FI-916195. The final numerical value for samples C and D can now be calculated using the samples in the window shown in Figure 3.

The median and linear operations used to calculate the numerical value are determined by the edge data as follows: 30 If the consistency check has given a new line sample C to be interpolated below sample c3, Figure 5a, the edge data is "left 4", ie the edge is negatively diagonal at 154 degrees. the values of pixels A and B as already mentioned above and the values of pixels C and D 35 are determined using a 3-point linear and median operation. The calculation formulas are then: 10 93295 left4: A = c3 B = 1/2 * (c3 + c4) C = Med3 (cl, p3, ell) D = 1/2 * [Med3 (cl, p3, ell) + Med3 ( C2, p4, Cl2)] 5

Depending on the direction of the edge given by the consistency check, the values for samples A, B, C and D are calculated as follows: left3: A = c3 10 B = 1/2 * (c3 + c4) C = 1/2 * [Med3 (cl, p2, clO) + Med3 (c2, p3, i.e.)] D = Med3 (c2, p3, i.e.) left2: A = c3 15 B = 1/2 * (c3 + c4) C = Med3 (c2, p3, clO) D = 1 / 2 * (Med3 (c2, p3, clO) + Med3 (c3, p4, i.e.)] left: A = c3 20 B = 1/2 * (c3 + c4) C = 1/2 * [Med3 (c2, p2) , c9) + Med3 (c3, p3, clO)] D = Med3 (c3, p3, ClO) vertical: A = c3 B = 1/2 * (c3 + c4) C = Med3 (c3, p3, c9) D = 1/2 * [Med3 (c3, p3, c9) + Med3 (c4, p4, ClO)] right: A = c3 30 B = 1/2 * (c3 + c4) C = 1/2 * [Med3 (c3, p2, c8) + Med3 (c4, p3, c9)] D = Med3 (c4, p3, c9) right2: A = c3 35 B = 1/2 * (c3 + c4) C = Med3 [c4, p3, c8] D = 1/2 * [Med3 (c4, p3, c8) + (Med3 (c5, p4, c9)] 11 93295 right3: A = c3 B = 1/2 * (c3 + c4) C = 1/2 * [Med3 (c4, p2, c7) + Med3 (c5, p3, c8)] D = Med3 (c5, p3, C8) 5 right4: A = c3 B = 1/2 * (c3 + c4) C = Med3 (c5, p3, cl) D = 1/2 * [Med3 (c5, p3, c7) + Med3 (c6, p4, c8)] 10

In the formulas listed above, Med3 () denotes a median three-point operation, where the output of the interpolator is the median of its three inputs. As can be seen, pixel B is always interpolated as an average of the known pixels c3 and c4 and is thus independent of the edge direction, while the value of pixels C and D is dependent on the edge direction.

As described above, the sample rate is increased regardless of whether the received field is even or odd. However, the smoothly moving motion of some smoothly moving image areas has become "motion judder" due to the fact that the values of the new samples do not fully correspond to their position in the HDTV rasteris-25 sa. Therefore, it is preferable to interpolate the new pixels of the received even and odd field differently. This highly preferred embodiment of the invention is described with reference to Figures 5a-c, in particular Figures 5b and 5c.

Assume that an odd field t of the normal format is received. Its sample density is increased just as shown above with reference to Fig. 5a, i.e. the sample values A (= C3), B, C and D are interpolated using the calculation formulas shown. When an even field is received (t-1 if 35 the previous even field is viewed, t + 1 if the next even field is viewed), the field samples cl ', c2' ..... cl2 'and the previous field samples p2', p3 'are calculated. and p4 'using the same window and the same interpolators 12 93295, the new samples A (= c3'), B, C and D exactly as shown in the description of Figure 5a. Now only the "dotted" pixels of the odd field t-1 are replaced by the "dotted" pixels of the even field t. The result is 5 situation 5 b situation. So far, progress has been made according to the first embodiment (Fig. 5a). Now, with the help of the interpolated pixels A, B, C and D of the even field t-1 and the pixels in the field of the field 6 * 3, further calculations are performed, resulting in the final new interpolated pixels E, F, G and H of the even field 10. .

This additional operation is illustrated by the transition arrow from Figure 5b to Figure 5c.

These new pixel values are calculated as follows: 15 E = (A + C) / 2 F = (B + D) / 2 G = (C + C9 ') / 2 H = (D + C9' + CIO ') / 3.

20 When these operations are performed on all pixels as the even field is received, a new HDTV field is formed in which all the pixels are interpolated and thus the original received pixels are not preserved. After all, in the odd HDTV field (Fig. 5a), the original pixels remained 25 (A = c3).

In the above, additional operations have been performed when receiving an even field, but it is equally possible to proceed in such a way that they are performed on an odd field and an even field is processed, as shown in connection with the description of Fig. 5a. This option is illustrated by the text inside the parentheses in the titles of Figures 5 a-c. It is therefore essential that the values of the new samples in every other field are corrected to better match their position in the HDTV raster-35 sa.

Thus, as described above, the edge information is first determined using the pixels of the current field, and according to the direction given by it, three new pixel values are calculated as the neighbor of the known pixel c3, also using the pixels of the previous field in the interpolation. Every other field corrects the new values obtained. When a video signal according to its traditional format is received, all its pixels are continuously subjected to the above-described operations, resulting in a field according to the HDTV format, which can be shown in Fig. 7. In it, dark squares denote original pixels and light squares denote interpolated 10 new pixels. The original PAL or similar field has thus quadrupled in sample density. The great advantage of the method is that the quadrupling of the sample density is done in one step. This is made possible by the use of the same 6 * 3 window for both edge detection and interpolation of new pixels.

Claims (12)

A method of doubling the sample density of a line burst scanned video signal arriving in sample sequence form, both horizontally and vertically, by interpolating on the same line a first new sample (A) and a second new sample (B ) for each known sample and on a line formed below a second (C) and third (D) new sample, characterized in that - at the same time, the line (m) of the field (t) is led to the first row of the window 6 * 3, the following line (m + 1) in the same field to the last row of said window and the line (m) of the preceding field (t-1) to be swiped between them to the middle row of the window, the top and bottom rows of the window are six successive samples cl, c2, c3, c4, c5, c6; c7, c8, c9, clO, eli, cl2 and in the middle row there are three successive samples p2, p3, p4 and where new samples are formed corresponding to the instantaneous sample on the top row, - using summation of the absolute difference signals. The lines in nine different directions of the sample pair on the top row of the window and the sample pair on the bottom row of the window determine the direction of any edges in the window (left 1, left 2, left 3, left 4, vertical, right, right2, right3, right4). 25. linear operation and 3-point median operation (M3) are interpolated into the window sample by applying a first (A), second (B), third (C) and fourth (D) new sample in the direction indicated by the edge. Method according to Claim 1, characterized in that for determining the edge information: - the samples are calculated on the top and bottom row of the window for each edge direction as follows: 90 ° = | c3 - c9 | + | c4 - cl0 | 64 ° = | c3 - c8 | + | C4 - c9 | 116 ° = | c2 - c9 | + | C3 - Cl0 | 135 ° = | c2 - cl0 | + | c3 - cll | 93295 146 ° = I cl - CIO I + | c2 - cll | 154 ° = I C1 - C11 | + | c2 - C12 | 45 ° = Ic4 - c8I + Ic5 - c9I 34 ° = Ic4 - c7 | + | c5 - c8I 5 26 ° = Ic5 - C7I + Ic6 - c8 | - new samples are interpolated in the direction the minimum resistance indicates. Method according to claim 2, characterized in that - the first new sample (A) is the same as the third new sample (c3) on the top row of the window and the second new sample (B) is the same as the average of said third sample - and when interpolating the third (C) and fourth (D) new samples, the products of the 3-point median interpolator are a known sample on each row of the window and the direction of interpolation is chosen depending on the edge direction (90 ° , 64 °, 116 °, 135 °, 146 °, 154 °, 45 °, 34 °, 26 °) as follows: 154 °: C = Med3 (cl, p3, or) D = 1/2 * [Med3 (cl, p3, C11) + Med3 (c2, p4, Cl2)] 146 °: C = 1/2 * [Med3 (cl, p2, clO) + Med3 (c2, p3, cll)] D = Med3 ( c2, p3, cl1) 135 °: C = Med3 (c2, p3, ClO) D = 1/2 * [Med3 (c2, p3, clO) + Med3 (c3, p4, cll)] 116 ° C: C = 1/2 * [Med3 (c2, p2, c9) + Med3 (c3, p3, clO)] D = Med3 (c3, p3, clO) 90 °: C = Med3 (c3, p3, c9) D = 1/2 * [Med3 (c3, p3, c9) + Med3 (c4, p4, c10)] 64 °: C = 1/2 * [Med3 (c3, p2, c8) + Med3 (c4, p3) , c9)] D = Med3 (c4, p3, c9) 45 °: C = Med3 [c4, p3, C8] D = 1/2 * [Med3 (c4, p3, c8) + (Med3 (c5, p4, c9)] 93295 34 °: C = 1/2 * [Med3 (c4, p2, c7) + Med3 (c5, p3, c8)] D = Med3 (c5, p3, c8) 26 °: C = Med3 (c5, p3, c7) D = 1/2 * [Med3 (c5, p3, c7) + Med3 (c6f p4, c8)] 10 4 Method according to claim 2, characterized in that the resistor is calculated for several samples, which surrounds the second new sample C to be interpolated and the resistor is subjected to consistency checking, the result of which is the final edge information. 15 Method according to claim 3, characterized in that upon receipt of every two faults - the first received new sample (A) is replaced by the average (A + C) / 2 thereof and the third new sample (C), 20. the second new sample (B) is replaced by the average (B + D) / 2 of this and the fourth new sample (D), - the received third new sample (C) is replaced by the average ((C + c9) / 2) and the third sample (c9) on the bottom row of the window; 25. the received fourth new sample (D) is replaced by the average ((D + c9 '+ clO') / 3) of this and the third (c9) and the fourth sample (clO) on the bottom row of the window. A coupling for doubling the sample density of an interlaced video signal arriving in sample sequence form at the same time, both horizontally and vertically, comprising: - a window into which the video signal lines are led and from which instantaneous samples can be simultaneously transmitted for further processing; calculating means for calculating the absolute difference signals in different directions, - interpolation means with which edge-adaptively interpolated for each known sample on the same line a first new sample 93295 (A) and a second new sample (B), and in the line formed below a second (C) and a third (D) new sample, characterized in that - the window is a 6 * 3 window, pi whose first row simultaneously contains six successive samples cl, c2, c3, c4, c5, c6 of the file. the line (ml) of the t (t), p in the bottom row there are six successive samples c7, c8, c9, clO, or, cl2 of the same felt (t) following line (m + 1) and in the middle row three samples p2, p3, p4 of it precedes nth field (t-1) line (m) to be swiped between said lines; - the calculating means includes circuits for summing the absolute difference signals of the sample pair on the top row of the window and the sample pair in the bottom row of the window in nine different directions, and other circuits coupled to said circuits 15 to determine the edge direction of the window, - interpolators connected to the calculating means lining operation and 3-point median operation on the window's sample interpolate new samples (A, B, C and D) in the determined edge direction. 20 Coupling according to claim 6, characterized in that in order to determine the edge information: - the calculating means calculates the resistance of each edge in sample of the top and bottom rows of the window as follows: 90 ° = Ic3 - c9I + Ic4 - clOI 64 ° = Ic3 - c8I + | c4 - c9I 116 ° = Ic2 - c9I + | c3 - clOI 135 ° = Ic2 - clOI + | c3 - cll | 146 ° = I cl - CIO I + | c2 - cll | 154 ° = | cl - cll | + Ic2 - cl2 | 45 ° = Ic4 - c8I + IC5 - c9I 34 ° = Ic4 - c7I + Ic5 - c8I 26 ° = | c5 - c7I + IC6 - C8I 35 - interpolators interpolate new samples in the direction the minimum resistor indicates. 93295 8. Coupling according to claim 6 or 7, characterized in that - as the first new sample (A), the same as the third sample (c3) on the top row of the window and as the second new sample (B) is stalled the average of said third new sample (c3) and subsequent sample (c4), - the products of the 3-point median operator, when interpolating the third new sample (C) and the fourth new sample (D), are a sample from each window of the window, The tiers depend on the edge direction (90 °, 64 °, 116 °, 135 °, 146 °, 154 °, 45 °, 34 °, 26 °) as follows: 154 °: C = Med3 (cl, p3, C11) D = 1/2 * [Med3 (cl, p3, cll) + Med3 (c2, p4, cl2)] 146 °: C = 1/2 * [Med3 (cl, p2, clO) + Med3 (c2, p3, c1)] D = Med3 (c2, p3, cl1) 135 °: C = Med3 (c2, p3, clO) D = 1/2 * [Med3 (c2, p3, ClO) + Med3 (c3, p4, cll)] 116 °: C = 1/2 * [Med3 (c2, p2, c9) + Med3 (c3, p3) , clO)] D = Med3 (c3, p3, clO) 90 °: C = Med3 (c3, p3, c9) D = 1/2 * [Med3 (c3, p3, c9) + Med3 (c4, p4, ClO)] 64 °: C = 1/2 * [Med3 (c3, p2, c8) + Med3 (c4, p3, c9)] D = Med3 (c4, p3, c9) 45 °: C = Med3 [c4 , p3, c8] D = 1/2 * [Med3 (c4, p3, c8) + (Med3 (c5, p4, c9)] 34 °: C = 1/2 * [Med3 (c4, p2, c7) + Med3 (c5, p3, c8)] D = Med3 (c5, p3, c8) 26 °: C = Med3 (c5, p3, C7) D = 1/2 * [Med3 (c5, p3, C7) + Med3 (c6, p4, c8)] II 93295 9. A coupling according to claim 8, characterized in that, upon receipt of every two fields, the interpolated new window samples are additionally passed to a means which calculates their average on the basis of their input values, whose output value replaces - the first new sample (A) where the products consists of a first new sample (A) and a third new sample (C), - the second new sample (B) since the products comprise a second new sample (B) and a fourth new sample (D), 10. the third new sample (C) where the products consist of said sample and the third sample (c9) on the bottom row of the window, - the fourth new sample (D) where the products consist of said sample (D) and the third (c9) and fourth sample ( clO) on the bottom row of the window. Coupling according to claim 7, characterized in that the calculating means comprise a memory in which the edge information regarding the known sample of the window is recorded, and a consistency control circuit which performs consistency checking for edge information read in the memory, the result of which is the definitive edge information.